[PATCH 4/8] drm/i915/gvt: Move common vGPU workload creation into scheduler.c
Zhi Wang
zhi.a.wang at intel.com
Wed Sep 13 18:50:44 UTC 2017
Move common vGPU workload creation functions into scheduler.c since
they are not specific to execlist emulation.
Signed-off-by: Zhi Wang <zhi.a.wang at intel.com>
---
drivers/gpu/drm/i915/gvt/execlist.c | 144 +--------------------------
drivers/gpu/drm/i915/gvt/scheduler.c | 183 ++++++++++++++++++++++++++++++++---
drivers/gpu/drm/i915/gvt/scheduler.h | 3 +-
3 files changed, 174 insertions(+), 156 deletions(-)
diff --git a/drivers/gpu/drm/i915/gvt/execlist.c b/drivers/gpu/drm/i915/gvt/execlist.c
index 6110837..22c7edf 100644
--- a/drivers/gpu/drm/i915/gvt/execlist.c
+++ b/drivers/gpu/drm/i915/gvt/execlist.c
@@ -443,169 +443,27 @@ static int complete_execlist_workload(struct intel_vgpu_workload *workload)
return ret;
}
-#define RING_CTX_OFF(x) \
- offsetof(struct execlist_ring_context, x)
-
-static void read_guest_pdps(struct intel_vgpu *vgpu,
- u64 ring_context_gpa, u32 pdp[8])
-{
- u64 gpa;
- int i;
-
- gpa = ring_context_gpa + RING_CTX_OFF(pdp3_UDW.val);
-
- for (i = 0; i < 8; i++)
- intel_gvt_hypervisor_read_gpa(vgpu,
- gpa + i * 8, &pdp[7 - i], 4);
-}
-
-static int prepare_mm(struct intel_vgpu_workload *workload)
-{
- struct execlist_ctx_descriptor_format *desc = &workload->ctx_desc;
- struct intel_vgpu_mm *mm;
- struct intel_vgpu *vgpu = workload->vgpu;
- int page_table_level;
- u32 pdp[8];
-
- if (desc->addressing_mode == 1) { /* legacy 32-bit */
- page_table_level = 3;
- } else if (desc->addressing_mode == 3) { /* legacy 64 bit */
- page_table_level = 4;
- } else {
- gvt_vgpu_err("Advanced Context mode(SVM) is not supported!\n");
- return -EINVAL;
- }
-
- read_guest_pdps(workload->vgpu, workload->ring_context_gpa, pdp);
-
- mm = intel_vgpu_find_ppgtt_mm(workload->vgpu, page_table_level, pdp);
- if (mm) {
- intel_gvt_mm_reference(mm);
- } else {
-
- mm = intel_vgpu_create_mm(workload->vgpu, INTEL_GVT_MM_PPGTT,
- pdp, page_table_level, 0);
- if (IS_ERR(mm)) {
- gvt_vgpu_err("fail to create mm object.\n");
- return PTR_ERR(mm);
- }
- }
- workload->shadow_mm = mm;
- return 0;
-}
-
-#define get_last_workload(q) \
- (list_empty(q) ? NULL : container_of(q->prev, \
- struct intel_vgpu_workload, list))
-
static int submit_context(struct intel_vgpu *vgpu, int ring_id,
struct execlist_ctx_descriptor_format *desc,
bool emulate_schedule_in)
{
struct intel_vgpu_submission *s = &vgpu->submission;
- struct list_head *q = workload_q_head(vgpu, ring_id);
- struct intel_vgpu_workload *last_workload = get_last_workload(q);
struct intel_vgpu_workload *workload = NULL;
- struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
- u64 ring_context_gpa;
- u32 head, tail, start, ctl, ctx_ctl, per_ctx, indirect_ctx;
- int ret;
-
- ring_context_gpa = intel_vgpu_gma_to_gpa(vgpu->gtt.ggtt_mm,
- (u32)((desc->lrca + 1) << GTT_PAGE_SHIFT));
- if (ring_context_gpa == INTEL_GVT_INVALID_ADDR) {
- gvt_vgpu_err("invalid guest context LRCA: %x\n", desc->lrca);
- return -EINVAL;
- }
-
- intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
- RING_CTX_OFF(ring_header.val), &head, 4);
-
- intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
- RING_CTX_OFF(ring_tail.val), &tail, 4);
-
- head &= RB_HEAD_OFF_MASK;
- tail &= RB_TAIL_OFF_MASK;
-
- if (last_workload && same_context(&last_workload->ctx_desc, desc)) {
- gvt_dbg_el("ring id %d cur workload == last\n", ring_id);
- gvt_dbg_el("ctx head %x real head %lx\n", head,
- last_workload->rb_tail);
- /*
- * cannot use guest context head pointer here,
- * as it might not be updated at this time
- */
- head = last_workload->rb_tail;
- }
-
- gvt_dbg_el("ring id %d begin a new workload\n", ring_id);
- /* record some ring buffer register values for scan and shadow */
- intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
- RING_CTX_OFF(rb_start.val), &start, 4);
- intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
- RING_CTX_OFF(rb_ctrl.val), &ctl, 4);
- intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
- RING_CTX_OFF(ctx_ctrl.val), &ctx_ctl, 4);
-
- workload = intel_vgpu_create_workload(vgpu);
+ workload = intel_vgpu_create_workload(vgpu, ring_id, desc);
if (IS_ERR(workload))
return PTR_ERR(workload);
- workload->ring_id = ring_id;
- workload->ctx_desc = *desc;
- workload->ring_context_gpa = ring_context_gpa;
- workload->rb_head = head;
- workload->rb_tail = tail;
- workload->rb_start = start;
- workload->rb_ctl = ctl;
workload->prepare = prepare_execlist_workload;
workload->complete = complete_execlist_workload;
workload->emulate_schedule_in = emulate_schedule_in;
- if (ring_id == RCS) {
- intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
- RING_CTX_OFF(bb_per_ctx_ptr.val), &per_ctx, 4);
- intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
- RING_CTX_OFF(rcs_indirect_ctx.val), &indirect_ctx, 4);
-
- workload->wa_ctx.indirect_ctx.guest_gma =
- indirect_ctx & INDIRECT_CTX_ADDR_MASK;
- workload->wa_ctx.indirect_ctx.size =
- (indirect_ctx & INDIRECT_CTX_SIZE_MASK) *
- CACHELINE_BYTES;
- workload->wa_ctx.per_ctx.guest_gma =
- per_ctx & PER_CTX_ADDR_MASK;
-
- WARN_ON(workload->wa_ctx.indirect_ctx.size && !(per_ctx & 0x1));
- }
-
if (emulate_schedule_in)
workload->elsp_dwords = s->execlist[ring_id].elsp_dwords;
- gvt_dbg_el("workload %p ring id %d head %x tail %x start %x ctl %x\n",
- workload, ring_id, head, tail, start, ctl);
-
gvt_dbg_el("workload %p emulate schedule_in %d\n", workload,
emulate_schedule_in);
- ret = prepare_mm(workload);
- if (ret) {
- kmem_cache_free(s->workloads, workload);
- return ret;
- }
-
- /* Only scan and shadow the first workload in the queue
- * as there is only one pre-allocated buf-obj for shadow.
- */
- if (list_empty(workload_q_head(vgpu, ring_id))) {
- intel_runtime_pm_get(dev_priv);
- mutex_lock(&dev_priv->drm.struct_mutex);
- intel_gvt_scan_and_shadow_workload(workload);
- mutex_unlock(&dev_priv->drm.struct_mutex);
- intel_runtime_pm_put(dev_priv);
- }
-
queue_workload(workload);
return 0;
}
diff --git a/drivers/gpu/drm/i915/gvt/scheduler.c b/drivers/gpu/drm/i915/gvt/scheduler.c
index 3eaa1df..544a887 100644
--- a/drivers/gpu/drm/i915/gvt/scheduler.c
+++ b/drivers/gpu/drm/i915/gvt/scheduler.c
@@ -953,9 +953,90 @@ void intel_vgpu_destroy_workload(struct intel_vgpu_workload *workload)
kmem_cache_free(s->workloads, workload);
}
+static struct intel_vgpu_workload *
+alloc_workload(struct intel_vgpu *vgpu)
+{
+ struct intel_vgpu_submission *s = &vgpu->submission;
+ struct intel_vgpu_workload *workload;
+
+ workload = kmem_cache_zalloc(s->workloads, GFP_KERNEL);
+ if (!workload)
+ return ERR_PTR(-ENOMEM);
+
+ INIT_LIST_HEAD(&workload->list);
+ INIT_LIST_HEAD(&workload->shadow_bb);
+
+ init_waitqueue_head(&workload->shadow_ctx_status_wq);
+ atomic_set(&workload->shadow_ctx_active, 0);
+
+ workload->status = -EINPROGRESS;
+ workload->shadowed = false;
+ workload->vgpu = vgpu;
+
+ return workload;
+}
+
+#define RING_CTX_OFF(x) \
+ offsetof(struct execlist_ring_context, x)
+
+static void read_guest_pdps(struct intel_vgpu *vgpu,
+ u64 ring_context_gpa, u32 pdp[8])
+{
+ u64 gpa;
+ int i;
+
+ gpa = ring_context_gpa + RING_CTX_OFF(pdp3_UDW.val);
+
+ for (i = 0; i < 8; i++)
+ intel_gvt_hypervisor_read_gpa(vgpu,
+ gpa + i * 8, &pdp[7 - i], 4);
+}
+
+static int prepare_mm(struct intel_vgpu_workload *workload)
+{
+ struct execlist_ctx_descriptor_format *desc = &workload->ctx_desc;
+ struct intel_vgpu_mm *mm;
+ struct intel_vgpu *vgpu = workload->vgpu;
+ int page_table_level;
+ u32 pdp[8];
+
+ if (desc->addressing_mode == 1) { /* legacy 32-bit */
+ page_table_level = 3;
+ } else if (desc->addressing_mode == 3) { /* legacy 64 bit */
+ page_table_level = 4;
+ } else {
+ gvt_vgpu_err("Advanced Context mode(SVM) is not supported!\n");
+ return -EINVAL;
+ }
+
+ read_guest_pdps(workload->vgpu, workload->ring_context_gpa, pdp);
+
+ mm = intel_vgpu_find_ppgtt_mm(workload->vgpu, page_table_level, pdp);
+ if (mm) {
+ intel_gvt_mm_reference(mm);
+ } else {
+
+ mm = intel_vgpu_create_mm(workload->vgpu, INTEL_GVT_MM_PPGTT,
+ pdp, page_table_level, 0);
+ if (IS_ERR(mm)) {
+ gvt_vgpu_err("fail to create mm object.\n");
+ return PTR_ERR(mm);
+ }
+ }
+ workload->shadow_mm = mm;
+ return 0;
+}
+
+#define same_context(a, b) (((a)->context_id == (b)->context_id) && \
+ ((a)->lrca == (b)->lrca))
+
+#define get_last_workload(q) \
+ (list_empty(q) ? NULL : container_of(q->prev, \
+ struct intel_vgpu_workload, list))
/**
* intel_vgpu_create_workload - create a vGPU workload
* @vgpu: a vGPU
+ * @desc: a guest context descriptor
*
* This function is called when creating a vGPU workload.
*
@@ -965,24 +1046,102 @@ void intel_vgpu_destroy_workload(struct intel_vgpu_workload *workload)
*
*/
struct intel_vgpu_workload *
-intel_vgpu_create_workload(struct intel_vgpu *vgpu)
+intel_vgpu_create_workload(struct intel_vgpu *vgpu, int ring_id,
+ struct execlist_ctx_descriptor_format *desc)
{
struct intel_vgpu_submission *s = &vgpu->submission;
- struct intel_vgpu_workload *workload;
+ struct list_head *q = workload_q_head(vgpu, ring_id);
+ struct intel_vgpu_workload *last_workload = get_last_workload(q);
+ struct intel_vgpu_workload *workload = NULL;
+ struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
+ u64 ring_context_gpa;
+ u32 head, tail, start, ctl, ctx_ctl, per_ctx, indirect_ctx;
+ int ret;
- workload = kmem_cache_zalloc(s->workloads, GFP_KERNEL);
- if (!workload)
- return ERR_PTR(-ENOMEM);
+ ring_context_gpa = intel_vgpu_gma_to_gpa(vgpu->gtt.ggtt_mm,
+ (u32)((desc->lrca + 1) << GTT_PAGE_SHIFT));
+ if (ring_context_gpa == INTEL_GVT_INVALID_ADDR) {
+ gvt_vgpu_err("invalid guest context LRCA: %x\n", desc->lrca);
+ return ERR_PTR(-EINVAL);
+ }
- INIT_LIST_HEAD(&workload->list);
- INIT_LIST_HEAD(&workload->shadow_bb);
+ intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
+ RING_CTX_OFF(ring_header.val), &head, 4);
- init_waitqueue_head(&workload->shadow_ctx_status_wq);
- atomic_set(&workload->shadow_ctx_active, 0);
+ intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
+ RING_CTX_OFF(ring_tail.val), &tail, 4);
- workload->status = -EINPROGRESS;
- workload->shadowed = false;
- workload->vgpu = vgpu;
+ head &= RB_HEAD_OFF_MASK;
+ tail &= RB_TAIL_OFF_MASK;
+
+ if (last_workload && same_context(&last_workload->ctx_desc, desc)) {
+ gvt_dbg_el("ring id %d cur workload == last\n", ring_id);
+ gvt_dbg_el("ctx head %x real head %lx\n", head,
+ last_workload->rb_tail);
+ /*
+ * cannot use guest context head pointer here,
+ * as it might not be updated at this time
+ */
+ head = last_workload->rb_tail;
+ }
+
+ gvt_dbg_el("ring id %d begin a new workload\n", ring_id);
+ /* record some ring buffer register values for scan and shadow */
+ intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
+ RING_CTX_OFF(rb_start.val), &start, 4);
+ intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
+ RING_CTX_OFF(rb_ctrl.val), &ctl, 4);
+ intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
+ RING_CTX_OFF(ctx_ctrl.val), &ctx_ctl, 4);
+
+ workload = alloc_workload(vgpu);
+ if (IS_ERR(workload))
+ return workload;
+
+ workload->ring_id = ring_id;
+ workload->ctx_desc = *desc;
+ workload->ring_context_gpa = ring_context_gpa;
+ workload->rb_head = head;
+ workload->rb_tail = tail;
+ workload->rb_start = start;
+ workload->rb_ctl = ctl;
+
+ if (ring_id == RCS) {
+ intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
+ RING_CTX_OFF(bb_per_ctx_ptr.val), &per_ctx, 4);
+ intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa +
+ RING_CTX_OFF(rcs_indirect_ctx.val), &indirect_ctx, 4);
+
+ workload->wa_ctx.indirect_ctx.guest_gma =
+ indirect_ctx & INDIRECT_CTX_ADDR_MASK;
+ workload->wa_ctx.indirect_ctx.size =
+ (indirect_ctx & INDIRECT_CTX_SIZE_MASK) *
+ CACHELINE_BYTES;
+ workload->wa_ctx.per_ctx.guest_gma =
+ per_ctx & PER_CTX_ADDR_MASK;
+
+ WARN_ON(workload->wa_ctx.indirect_ctx.size && !(per_ctx & 0x1));
+ }
+
+ gvt_dbg_el("workload %p ring id %d head %x tail %x start %x ctl %x\n",
+ workload, ring_id, head, tail, start, ctl);
+
+ ret = prepare_mm(workload);
+ if (ret) {
+ kmem_cache_free(s->workloads, workload);
+ return ERR_PTR(ret);
+ }
+
+ /* Only scan and shadow the first workload in the queue
+ * as there is only one pre-allocated buf-obj for shadow.
+ */
+ if (list_empty(workload_q_head(vgpu, ring_id))) {
+ intel_runtime_pm_get(dev_priv);
+ mutex_lock(&dev_priv->drm.struct_mutex);
+ intel_gvt_scan_and_shadow_workload(workload);
+ mutex_unlock(&dev_priv->drm.struct_mutex);
+ intel_runtime_pm_put(dev_priv);
+ }
return workload;
}
diff --git a/drivers/gpu/drm/i915/gvt/scheduler.h b/drivers/gpu/drm/i915/gvt/scheduler.h
index 9435d7e..9f4ff8f 100644
--- a/drivers/gpu/drm/i915/gvt/scheduler.h
+++ b/drivers/gpu/drm/i915/gvt/scheduler.h
@@ -141,7 +141,8 @@ int intel_vgpu_setup_submission(struct intel_vgpu *vgpu);
void intel_vgpu_clean_submission(struct intel_vgpu *vgpu);
struct intel_vgpu_workload *
-intel_vgpu_create_workload(struct intel_vgpu *vgpu);
+intel_vgpu_create_workload(struct intel_vgpu *vgpu, int ring_id,
+ struct execlist_ctx_descriptor_format *desc);
void intel_vgpu_destroy_workload(struct intel_vgpu_workload *workload);
--
2.7.4
More information about the intel-gvt-dev
mailing list